Agents to assist in finding help

Abstract

The problem of finding someone who might be able to help with a particular task or knowledge area exists everywhere, be it in groups of students or corporate settings. Time and effort are spent looking for relevant information when another person in the community could easily provide assistance. Our approach to addressing this problem is to use software agents to assist the search for expertise. Previous research on this topic has been mostly in the areas of information marketplaces, referral systems or information repositories. We built a system, called Expert Finder, which provides a testbed for ideas and techniques developed in the context of this thesis. Expert Finder analyzes previous work of both the novice and the expert to automatically categorize expertise and match it with the user's problem while providing a community-based incentive mechanism. We present alternative profiling and incentive mechanisms to those that had been presented in previous work. We chose the Java Programming domain for our initial implementation and testing of the system. Expert Finder uses each user's Java source files to determine their expertise and uses a Java domain model to match questions and experts. It keeps track of users' willingness to help out as a measure of the social capital in the community. We ran some tests using the prototype system to assess how well these ideas worked, and results are also reported in this thesis.